Gregory Bateson, in Mind and Nature, writes:
To liken the mountain to a man and talk of its “humor” or “rage” does little harm. But to liken the man to the mountain proposes that all human relationships are what Martin Buber might have called I-it or perhaps it-it relations. The mountain, personified in our speech, will not become a person, will not learn a more personal way of being. But the human being, depersonified in his own talk, and thought, may indeed learn more thingish habits of action.
Bateson 1979:101
Two thoughts on this.
First thought. I haven’t edited out the casual sexism here because it is fitting to the case this made me think of: OpenAI CEO Sam Altman proudly declaring “I am a stochastic parrot”, appropriating the metaphor Emily Bender and colleagues put forward to highlight the nature of large language models. It was a cheap shot, and (therefore) it was widely boosted by techbros, who have long given up about reading or understanding any sophisticated argument, and especially those that may put cracks in their visions of a gleaming white android future (on which see Erscoi et al., Pygmalion displacement).
It fits a larger pattern where people are happy to not think deeply anymore and go for the most superficial take. What Bateson points out is that this depersonification has consequences. One of the consequences, I think, is a shirking of accountability. By declaring himself a stochastic parrot, Sam Altman also says that we don’t need to try and hold him accountable for anything that he might say (or rather parrot). Neat trick for a soulless billionaire who believes he couldn’t raise a child without ChatGPT.
Second thought. Bateson’s example of the mountain personified is indeed mostly harmless. But even if metaphors applied to things have no direct effect on those, they do have an effect on our construals of them. Today, personification of LLMs is almost as common as depersonification of people. Just as we may “learn more thingish habits of action” by depersonifying ourselves, so we may ascribe to LLMs “more humanish properties” by personifying them. This is one of the risks of anthropomorphizing technology: ascription follows description (as I wrote in my ‘Interactional foundations’ paper). Or as Lucy Suchman (1987/2007) put it: “as soon as computational artifacts show evidence of some evidence of recognizably human abilities, we are inclined to endow them with the rest.”
- Bateson, Gregory. 1979. Mind and Nature. New York: E.P. Dutton.
- Bender, Emily M., Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. 2021. ‘On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜’. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 610–23. Virtual Event Canada: ACM. doi:10.1145/3442188.3445922.
- Dingemanse, M. (2026) Interactional foundations for critical AI literacies. In Ana Valdivia (ed.) A Research Agenda for Critical AI. Edward Elgar Publishing. (doi, pdf)
- Erscoi, Lelia, Annelies Kleinherenbrink, and Olivia Guest. 2023. ‘Pygmalion Displacement: When Humanising AI Dehumanises Women’. SocArXiv. doi:10.31235/osf.io/jqxb6.
- Suchman, Lucy A. 2007. Human-Machine Reconfigurations: Plans and Situated Actions. 2nd ed. Cambridge ; New York: Cambridge University Press.